TL;DR
Microsoft unveiled Project Solara at Build 2026, a cloud-on-a-chip platform for “agent-first devices” that run artificial intelligence agents instead of traditional apps. The two concept devices, a wearable badge and desktop companion, are being tested with Best Buy, CVS Health, Levi’s and Target.
At Build 2026, Microsoft unveiled Project Solara, a new cloud-on-a-chip platform designed from the ground up for devices that run AI agents instead of traditional applications. The platform includes a lightweight operating system built on top of AOSP, enterprise-grade security and management through Intune and Entra ID, and what Microsoft “calls” it.just-in-time UI,” the ability to dynamically adapt the interface of agent experiences regardless of what they are running. Two conceptual device reference designs were shown: wearable badge and desktop companionboth target enterprise workers.
The announcement is significant because it represents Microsoft’s first attempt to build an operating system and hardware platform around the premise of replacing apps with agents as the primary way people interact with computers. Google, Salesforce, and OpenAI are all building agent platformsbut Microsoft is the first company to extend the concept to purpose-built devices that are neither phones, PCs, nor tablets.
What the devices look like
The badge concept reimagines the corporate access badge as an always-connected AI companion. It includes a touchscreen, a fingerprint sensor for Hello for Business authentication, a far-field microphone array for voice interaction and a speaker, a side camera, and WiFi, Bluetooth, 5G, and satellite connectivity powered by Qualcomm wearable silicon. Wearing it, a nurse, retail associate, or office worker can preview upcoming meetings, tap to record an in-person conversation with a full transcription, or ask their agents a silent question.
The table concept is a small stationary device with a touch screen, dual microphone array, speaker, UWB presence sensor and MediaTek IoT silicon. It authenticates via facial recognition (Hello for Business) and provides AI agents access to the environment while the user is working. Connected to an external display via USB-C, it becomes a Windows 365 cloud PC client, giving enterprises a single device that serves as both an agent companion and a thin client.
Both devices are clearly not designed to run traditional apps. No app store, no browser-first experience, no traditional desktop. It assumes the entire interaction model the user’s interaction with the software is mediated by agents rather than by opening and navigating individual applications.
Platform architecture
Project Solara runs on MDEP (Microsoft Device Ecosystem Platform), an enterprise-grade operating system built on the Android Open Source Project. This is notable: Microsoft is building its next-generation device platform on Android’s open-source base rather than Windows, a pragmatic choice that gives it access to Android’s hardware compatibility and driver ecosystem while also allowing Microsoft to overlay its own agent shell, security model, and management stack.
The platform is built on three pillars. First, enterprise readiness: physical privacy controls, including Intune device management, Sign-in ID authentication, Hello for Business biometrics, and a hardware microphone mute button. Second, an agent-driven interaction model with real-time UI that adapts across different screen sizes, form factors, and access modes. Third, extensibility for multiple agents, both Microsoft’s own (Copilot, Researcher, Facilitator, new Priority Agent) and third-party agents built on Microsoft 365 Agents SDK, Copilot Studio or Microsoft Agent Framework.
Enterprise AI agents are now deployed in retail, financial services and healthcare, but they work on existing devices. Project Solara’s thesis is that purpose-built hardware built around how agents work, rather than how applications work, can deliver better experiences in specific workflows and environments.
Just in time UI
The most technically ambitious element of the ad is the real-time UI. Traditionally, each new device form factor requires developers to redesign their apps for the new screen size, resolution, and input method. This is one of the reasons why new device categories are expensive to create and struggle without a strong app ecosystem.
Microsoft’s response is that agents must create their own interfaces. On the small tab screen, the agent can show a minimal card with one move. On a desktop device, the same agent creates a richer visual layout. It creates a complete dashboard on the connected display. The agent adapts its presentation to the device rather than requiring developers to build a separate experience for each form factor.
Today, this works with semi-structured approaches such as adaptive cards. As AI models improve in creating layouts and interfacesMicrosoft expects the system to move towards an increasingly dynamic and eventually fully generative UI. The company makes it clear that full generative UI is “not here yet,” but is investing in the middle of the spectrum between responsive design and unlimited generation.
Who is testing
Hundreds of Microsoft employees are already using concept devices internally. The company also announced a private pilot program spanning retail, healthcare and consumer services with AccuWeather, Best Buy, CVS Health, Levi’s and Target. GitHub Copilot and Dragon Copilot (Microsoft’s healthcare AI) are both exploring agent-first experiences on the platform.
The enterprise agent AI market is rapidly consolidatingand Microsoft is betting that the next competitive advantage is not just having the best agents, but also delivering them through the most appropriate devices. A nurse wearing a Solara badge that captures patient interactions, pulls up relevant notes, and tracks follow-up tasks is a fundamentally different value proposition than the same nurse typing on a laptop between patients.
The key question is whether enterprises will adopt yet another category of devices, with all the procurement, management and change management involved. Microsoft’s answer is that agents reduce the cost of specialization: because the agent adapts to the device, not the other way around, the barrier to creating new form factors is reduced. The platform is designed to make it possible, not inevitable. Pilot partners will determine if this is also desirable.
